Characterizing medical patients with delirium: A cohort study comparing ICD-10 codes and a validated chart review method.

BACKGROUND: Delirium is a major cause of preventable mortality and morbidity in hospitalized adults, but accurately determining rates of delirium remains a challenge. OBJECTIVE: To characterize and compare medical inpatients identified as having delirium using two common methods, administrative data and retrospective chart review. METHODS: We conducted a retrospective study of 3881 randomly selected internal medicine hospital admissions from six acute care hospitals in Toronto and Mississauga, Ontario, Canada. Delirium status was determined using ICD-10-CA codes from hospital administrative data and through a previously validated chart review method. Baseline sociodemographic and clinical characteristics, processes of care and outcomes were compared across those without delirium in hospital and those with delirium as determined by administrative data and chart review. RESULTS: Delirium was identified in 6.3% of admissions by ICD-10-CA codes compared to 25.7% by chart review. Using chart review as the reference standard, ICD-10-CA codes for delirium had sensitivity 24.1% (95%CI: 21.5-26.8%), specificity 99.8% (95%CI: 99.5-99.9%), positive predictive value 97.6% (95%CI: 94.6-98.9%), and negative predictive value 79.2% (95%CI: 78.6-79.7%). Age over 80, male gender, and Charlson comorbidity index greater than 2 were associated with misclassification of delirium. Inpatient mortality and median costs of care were greater in patients determined to have delirium by ICD-10-CA codes (5.8% greater mortality, 95% CI: 2.0-9.5 and $6824 greater cost, 95%CI: 4713-9264) and by chart review (11.9% greater mortality, 95%CI: 9.5-14.2% and $4967 greater cost, 95%CI: 4415-5701), compared to patients without delirium. CONCLUSIONS: Administrative data are specific but highly insensitive, missing most cases of delirium in hospital. Mortality and costs of care were greater for both the delirium cases that were detected and missed by administrative data. Better methods of routinely measuring delirium in hospital are needed.

Young adults with a 22q11.2 microdeletion and the cost of aging with complexity in a population-based context.

PURPOSE: Information about the impact on the adult health care system is limited for complex rare pediatric diseases, despite their increasing collective prevalence that has paralleled advances in clinical care of children. Within a population-based health care context, we examined costs and multimorbidity in adults with an exemplar of contemporary genetic diagnostics. METHODS: We estimated direct health care costs over an 18-year period for adults with molecularly confirmed 22q11.2 microdeletion (cases) and matched controls (total 60,459 person-years of data) by linking the case cohort to health administrative data for the Ontario population (∼15 million people). We used linear regression to compare the relative ratio (RR) of costs and to identify baseline predictors of higher costs. RESULTS: Total adult (age ≥ 18) health care costs were significantly higher for cases compared with population-based (RR 8.5, 95% CI 6.5-11.1) controls, and involved all health care sectors. At study end, when median age was <30 years, case costs were comparable to population-based individuals aged 72 years, likelihood of being within the top 1st percentile of health care costs for the entire (any age) population was significantly greater for cases than controls (odds ratio [OR], for adults 17.90, 95% CI 7.43-43.14), and just 8 (2.19%) cases had a multimorbidity score of zero (vs 1483 (40.63%) controls). The 22q11.2 microdeletion was a significant predictor of higher overall health care costs after adjustment for baseline variables (RR 6.9, 95% CI 4.6-10.5). CONCLUSION: The findings support the possible extension of integrative models of complex care used in pediatrics to adult medicine and the potential value of genetic diagnostics in adult clinical medicine.

Characterizing the burden of biliary tract cancers across 28 hospitals in Ontario, Canada.

BACKGROUND AND AIMS: The incidence of biliary tract cancers (BTC) appears to be increasing worldwide. We analyzed the characteristics of BTC-related hospitalizations under medical services across 28 hospitals in Ontario, Canada. METHODS: This study uses data collected by GEMINI, a hospital research data network. BTC-related hospitalizations from 2015 to 2021 under the Department of Medicine or intensive care unit were captured using the International Classification of Diseases, 10th revision, codes for intrahepatic cholangiocarcinoma (iCCA), extrahepatic cholangiocarcinoma, and gallbladder cancers. RESULTS: A total of 4596 BTC-related hospitalizations (2720 iCCA, 1269 extrahepatic cholangiocarcinoma, 607 gallbladder cancers) were analyzed. The number of unique patients with BTC-related hospitalizations increased over time. For iCCA-related hospitalizations, the total number of hospitalizations increased (from 385 in 2016 to 420 in 2021, p = .005), the hospital length of stay decreased over the study period (mean 10 days [SD, 12] in 2016 to 9 days [SD, 8] in 2021, p = .04), and the number of in-hospital deaths was stable (from 68 [18%] in 2016 to 55 [13%] in 2021, p = .62). Other outcomes such as 30-day readmissions, medical imaging tests, intensive care unit-specific hospitalizations, and length of stay were stable over time for all cohorts. The cost of hospitalization for the BTC cohort increased from median $8203 CAD (interquartile range, 5063-15,543) in 2017 to $8507 CAD (interquartile range, 5345-14,755) in 2021. CONCLUSIONS: This real-world data analysis showed a rising number of patients with BTC-related hospitalizations and rising number of iCCA-related hospitalizations across 28 hospitals in Ontario between 2015 and 2021.

Decentralised, collaborative, and privacy-preserving machine learning for multi-hospital data.

BACKGROUND: Machine Learning (ML) has demonstrated its great potential on medical data analysis. Large datasets collected from diverse sources and settings are essential for ML models in healthcare to achieve better accuracy and generalizability. Sharing data across different healthcare institutions or jurisdictions is challenging because of complex and varying privacy and regulatory requirements. Hence, it is hard but crucial to allow multiple parties to collaboratively train an ML model leveraging the private datasets available at each party without the need for direct sharing of those datasets or compromising the privacy of the datasets through collaboration. METHODS: In this paper, we address this challenge by proposing Decentralized, Collaborative, and Privacy-preserving ML for Multi-Hospital Data (DeCaPH). This framework offers the following key benefits: (1) it allows different parties to collaboratively train an ML model without transferring their private datasets (i.e., no data centralization); (2) it safeguards patients' privacy by limiting the potential privacy leakage arising from any contents shared across the parties during the training process; and (3) it facilitates the ML model training without relying on a centralized party/server. FINDINGS: We demonstrate the generalizability and power of DeCaPH on three distinct tasks using real-world distributed medical datasets: patient mortality prediction using electronic health records, cell-type classification using single-cell human genomes, and pathology identification using chest radiology images. The ML models trained with DeCaPH framework have less than 3.2% drop in model performance comparing to those trained by the non-privacy-preserving collaborative framework. Meanwhile, the average vulnerability to privacy attacks of the models trained with DeCaPH decreased by up to 16%. In addition, models trained with our DeCaPH framework achieve better performance than those models trained solely with the private datasets from individual parties without collaboration and those trained with the previous privacy-preserving collaborative training framework under the same privacy guarantee by up to 70% and 18.2% respectively. INTERPRETATION: We demonstrate that the ML models trained with DeCaPH framework have an improved utility-privacy trade-off, showing DeCaPH enables the models to have good performance while preserving the privacy of the training data points. In addition, the ML models trained with DeCaPH framework in general outperform those trained solely with the private datasets from individual parties, showing that DeCaPH enhances the model generalizability. FUNDING: This work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC, RGPIN-2020-06189 and DGECR-2020-00294), Canadian Institute for Advanced Research (CIFAR) AI Catalyst Grants, CIFAR AI Chair programs, Temerty Professor of AI Research and Education in Medicine, University of Toronto, Amazon, Apple, DARPA through the GARD project, Intel, Meta, the Ontario Early Researcher Award, and the Sloan Foundation. Resources used in preparing this research were provided, in part, by the Province of Ontario, the Government of Canada through CIFAR, and companies sponsoring the Vector Institute.

Anaerobic Antibiotic Coverage in Aspiration Pneumonia and the Associated Benefits and Harms: A Retrospective Cohort Study.

BACKGROUND: Antibiotics with extended anaerobic coverage are used commonly to treat aspiration pneumonia, which is not recommended by current guidelines. RESEARCH QUESTION: In patients admitted to hospital for community-acquired aspiration pneumonia, does a difference exist between antibiotic therapy with limited anaerobic coverage (LAC) vs antibiotic therapy with extended anaerobic coverage (EAC) in terms of in-hospital mortality and risk of Clostridioides difficile colitis? STUDY DESIGN AND METHODS: We conducted a multicenter retrospective cohort study across 18 hospitals in Ontario, Canada, from January 1, 2015, to January 1, 2022. Patients were included if the physician diagnosed aspiration pneumonia and prescribed guideline-concordant first-line community-acquired pneumonia parenteral antibiotic therapy to the patient within 48 h of admission. Patients then were categorized into the LAC group if they received ceftriaxone, cefotaxime, or levofloxacin. Patients were categorized into the EAC group if they received amoxicillin-clavulanate, moxifloxacin, or any of ceftriaxone, cefotaxime, or levofloxacin in combination with clindamycin or metronidazole. The primary outcome was all-cause in-hospital mortality. Secondary outcomes included incident C difficile colitis occurring after admission. Overlap weighting of propensity scores was used to balance baseline prognostic factors. RESULTS: The LAC and EAC groups included 2,683 and 1,316 patients, respectively. In hospital, 814 patients (30.3%) and 422 patients (32.1%) in the LAC and EAC groups died, respectively. C difficile colitis occurred in 5 or fewer patients (≤ 0.2%) and 11 to 15 patients (0.8%-1.1%) in the LAC and EAC groups, respectively. After overlap weighting of propensity scores, the adjusted risk difference of EAC minus LAC was 1.6% (95% CI, -1.7% to 4.9%) for in-hospital mortality and 1.0% (95% CI, 0.3%-1.7%) for C difficile colitis. INTERPRETATION: Extended anaerobic coverage likely is unnecessary in aspiration pneumonia because it is associated with no additional mortality benefit, only an increased risk of C difficile colitis.

Advanced imaging use and delays among inpatients with psychiatric comorbidity.

OBJECTIVE: To determine whether presence of a psychiatric comorbidity impacts use of inpatient imaging tests and subsequent wait times. METHODS: This was a retrospective cohort study of all patients admitted to General Internal Medicine (GIM) at five academic hospitals in Toronto, Ontario from 2010 to 2019. Exposure was presence of a coded psychiatric comorbidity on admission. Primary outcome was time to test, as calculated from the time of test ordering to time of test completion, for computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, or peripherally inserted central catheter (PICC) insertion. Multilevel mixed-effects models were used to identify predictors of time to test, and marginal effects were used to calculate differences in absolute units (h). Secondary outcome was the rate of each type of test included. Subgroup analyses were performed according to type of psychiatric comorbidity: psychotic, mood/anxiety, or substance use disorder. RESULTS: There were 196,819 GIM admissions from 2010to 2019. In 77,562 admissions, ≥1 advanced imaging test was performed. After adjusting for all covariates, presence of any psychiatric comorbidity was associated with increased time to test for MRI (adjusted difference: 5.3 h, 95% confidence interval [CI]: 3.9-6.8), PICC (adjusted difference: 3.7 h, 95% CI: 1.6-5.8), and ultrasound (adjusted difference: 3.0 h, 95% CI: 2.3-3.8), but not for CT (adjusted difference: 0.1 h, 95% CI: -0.3 to 0.5). Presence of any psychiatric comorbidity was associated with lower rate of ordering for all test types (adjusted difference: -17.2 tests per 100 days hospitalization, interquartile range: -18.0 to -16.3). CONCLUSIONS: There was a lower rate of ordering of advanced imaging among patients with psychiatric comorbidity. Once ordered, time to test completion was longer for MRI, ultrasound, and PICC. Further exploration, such as quantifying rates of cancelled tests and qualitative studies evaluating hospital, provider, and patient barriers to timely advanced imaging, will be helpful in elucidating causes for these disparities.

Validation of the Diagnostic Accuracy Levels of International Classification of Diseases, 10th Revision Codes for Diabetic Ketoacidosis: A Multicentre, Cross-sectional Study of Adults.

OBJECTIVES: International Classification of Diseases (ICD) codes are commonly used to identify cases of diabetic ketoacidosis (DKA) in health services research, but they have not been validated. Our aim in this study was to assess the accuracy of ICD, 10th revision (ICD-10) diagnosis codes for DKA. METHODS: We conducted a multicentre, cross-sectional study using data from 5 hospitals in Ontario, Canada. Each hospitalization event has a single most responsible diagnosis code. We identified all hospitalizations assigned diagnosis codes for DKA. A true case of DKA was defined using laboratory values (serum bicarbonate ≤18 mmol/L, arterial pH ≤7.3, anion gap ≥14 mEq/L, and presence of ketones in urine or blood). Chart review was conducted to validate DKA if laboratory values were missing or the diagnosis of DKA was unclear. Outcome measures included positive predictive value (PPV), negative predictive value (NPV), sensitivity, and specificity of ICD-10 codes in patients with laboratory-defined DKA. RESULTS: We identified 316,517 hospitalizations. Among these, 312,948 did not have an ICD-10 diagnosis code for DKA and 3,569 had an ICD-10 diagnosis code for DKA. Using a combination of laboratory and chart review, we identified that the overall PPV was 67.0%, the NPV was 99.7%, specificity was 99.6%, and sensitivity was 74.9%. When we restricted our analysis to hospitalizations in which DKA was the most responsible discharge diagnosis (n=3,374 [94.5%]), the test characteristics were PPV 69.8%, NPV 99.7%, specificity 99.7%, and sensitivity 71.9%. CONCLUSION: ICD-10 codes can identify patients with DKA among those admitted to general internal medicine.

Comparative Effectiveness of First-Line and Alternative Antibiotic Regimens in Hospitalized Patients With Nonsevere Community-Acquired Pneumonia: A Multicenter Retrospective Cohort Study.

BACKGROUND: There are several antibiotic regimens to treat community-acquired pneumonia (CAP). RESEARCH QUESTION: In patients hospitalized to a non-ICU ward setting with CAP, is there a difference between first-line and alternative antibiotic regimens (ß-lactam plus macrolide [BL+M], ß-lactam [BL] alone, respiratory fluoroquinolone [FQ], or ß-lactam plus doxycycline [BL+D]) in terms of in-hospital mortality? STUDY DESIGN AND METHODS: This retrospective cohort study included consecutive patients admitted with CAP at 19 Canadian hospitals from 2015 to 2021. Taking a target trial approach, patients were categorized into the four antibiotic groups based on the initial antibiotic treatment within 48 h of admission. Patients with severe CAP requiring ICU admission in the first 48 h were excluded. The primary outcome was all-cause in-hospital mortality. Secondary outcome included time to being discharged alive. Propensity score and overlap weighting were used to balance covariates. RESULTS: Of 23,512 patients, 9,340 patients (39.7%) received BL+M, 9,146 (38.9%) received BL, 4,510 (19.2%) received FQ, and 516 (2.2%) received BL+D. The number of in-hospital deaths was 703 (7.5%) for the BL+M group, 888 (9.7%) for the BL group, 302 (6.7%) for the FQ group, and 31 (6.0%) for the BL+D group. The adjusted risk difference for in-hospital mortality when compared with BL+M was 1.5% (95% CI, -0.3% to 3.3%) for BL, -0.9% (95% CI, -2.9% to 1.1%) for FQ, and -1.9% (95% CI, -4.8% to 0.9%) for BL+D. Compared with BL+M, the subdistribution hazard ratio for being discharged alive was 0.90 (95% CI, 0.84-0.96) for BL, 1.07 (95% CI, 0.99-1.16) for FQ, and 1.04 (95% CI, 0.93-1.17) for BL+D. INTERPRETATION: BL+M, FQ, and BL+D had similar outcomes and can be considered effective regimens for nonsevere CAP. Compared with BL+M, BL was associated with longer time to discharge and the CI for mortality cannot exclude a small but clinically important increase in risk.

Grand rounds in methodology: key considerations for implementing machine learning solutions in quality improvement initiatives.

Machine learning (ML) solutions are increasingly entering healthcare. They are complex, sociotechnical systems that include data inputs, ML models, technical infrastructure and human interactions. They have promise for improving care across a wide range of clinical applications but if poorly implemented, they may disrupt clinical workflows, exacerbate inequities in care and harm patients. Many aspects of ML solutions are similar to other digital technologies, which have well-established approaches to implementation. However, ML applications present distinct implementation challenges, given that their predictions are often complex and difficult to understand, they can be influenced by biases in the data sets used to develop them, and their impacts on human behaviour are poorly understood. This manuscript summarises the current state of knowledge about implementing ML solutions in clinical care and offers practical guidance for implementation. We propose three overarching questions for potential users to consider when deploying ML solutions in clinical care: (1) Is a clinical or operational problem likely to be addressed by an ML solution? (2) How can an ML solution be evaluated to determine its readiness for deployment? (3) How can an ML solution be deployed and maintained optimally? The Quality Improvement community has an essential role to play in ensuring that ML solutions are translated into clinical practice safely, effectively, and ethically.

Validity of hospital diagnostic codes to identify SARS-CoV-2 infections in reference to polymerase chain reaction results: a descriptive study.

BACKGROUND: In 2020, International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes were created for laboratory-confirmed SARS-CoV-2 infections. We assessed the operating characteristics of ICD-10 discharge diagnostic code U07.1 within the General Medicine Inpatient Initiative (GEMINI). METHODS: GEMINI assembles hospitalization data (including administrative ICD-10 discharge diagnostic codes, laboratory results and demographic data) from hospitals in Ontario, Canada. We studied adults (age ≥ 18 yr) admitted during 2020 and tested at least once for SARS-CoV-2 via polymerase chain reaction (PCR) during (or within 48 h before) hospitalization. With PCR results as the reference standard, we calculated sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for ICD-10 code U07.1 hospital discharge diagnostic codes. Analyses were stratified by demographic data, calendar period and timing of the first test (within or after 48 h of hospital admission). RESULTS: In 11 852 hospitalizations with at least 1 SARS-CoV-2 PCR test, 444 (3.7%) were positive. The sensitivity of code U07.1 to identify SARS-CoV-2 infection was 97.8%, specificity was 99.5%, PPV was 88.2% and NPV was 99.9%. Operating characteristics were similar in most stratified analyses, but the specificity and PPV were lower if the first SARS-CoV-2 test was done more than 48 hours after admission. INTERPRETATION: The sensitivity, specificity, PPV and NPV of code U07.1 were high. This supports using code U07.1 to identify SARS-CoV-2 infection in hospitalization data.

Ceftriaxone and the Risk of Ventricular Arrhythmia, Cardiac Arrest, and Death Among Patients Receiving Lansoprazole.

Importance: The combination of ceftriaxone and lansoprazole has been shown to prolong the corrected QT interval on electrocardiogram. However, it is unknown whether this translates to clinically important patient outcomes. Objective: To compare lansoprazole with another proton pump inhibitor (PPI) during ceftriaxone treatment in terms of risk for ventricular arrhythmia, cardiac arrest, and in-hospital mortality. Design, Setting, and Participants: A retrospective cohort study including adult medical inpatients receiving ceftriaxone with lansoprazole or another PPI in 13 hospitals in Ontario, Canada, was conducted from January 1, 2015, to December 31, 2021. Exposure: Lansoprazole during ceftriaxone treatment vs other PPIs during ceftriaxone treatment. Main Outcomes and Measures: The primary outcome was a composite of ventricular arrhythmia or cardiac arrest that occurred after hospital admission. The secondary outcome was all-cause in-hospital mortality. Propensity-score weighting was used to adjust for covariates including hospital site, demographic characteristics, comorbidities, risk factors for ventricular arrhythmia, illness severity, admitting diagnoses, and concomitant medications. Results: Of the 31 152 patients hospitalized on internal medicine wards who were treated with ceftriaxone while receiving a PPI, 16 135 patients (51.8%) were male, and the mean (SD) age was 71.7 (16.0) years. The study included 3747 patients in the lansoprazole group and 27 405 patients in the other PPI group. Ventricular arrhythmia or cardiac arrest occurred in 126 patients (3.4%) within the lansoprazole group and 319 patients (1.2%) within the other PPI group. In-hospital mortality occurred in 746 patients (19.9%) within the lansoprazole group and 2762 patients (10.1%) in the other PPI group. After weighting using propensity scores, the adjusted risk difference for the lansoprazole group minus other PPI group was 1.7% (95% CI, 1.1%-2.3%) for ventricular arrhythmia or cardiac arrest and 7.4% (95% CI, 6.1%-8.8%) for in-hospital mortality. Conclusions and Relevance: The findings of this cohort study suggest that combination therapy with lansoprazole and ceftriaxone should be avoided. More studies are needed to determine whether these findings could be replicated in other populations and settings.

Identifying clusters of coexisting conditions and outcomes among adults admitted to hospital with community-acquired pneumonia: a multicentre cohort study.

BACKGROUND: Little is known about patterns of coexisting conditions and their influence on clinical care or outcomes in adults admitted to hospital for community-acquired pneumonia (CAP). We sought to evaluate how coexisting conditions cluster in this population to advance understanding of how multimorbidity affects CAP. METHODS: We studied 11 085 adults admitted to hospital with CAP at 7 hospitals in Ontario, Canada. Using cluster analysis, we identified patient subgroups based on clustering of comorbidities in the Charlson Comorbidity Index. We derived and replicated cluster analyses in independent cohorts (derivation sample 2010-2015, replication sample 2015-2017), then combined these into a total cohort for final cluster analyses. We described differences in medications, imaging and outcomes. RESULTS: Patients clustered into 7 subgroups. The low comorbidity subgroup (n = 3052, 27.5%) had no comorbidities. The DM-HF-Pulm subgroup had prevalent diabetes, heart failure and chronic lung disease (n = 1710, 15.4%). One disease category defined each remaining subgroup, as follows: pulmonary (n = 1621, 14.6%), diabetes (n = 1281, 11.6%), heart failure (n = 1370, 12.4%), dementia (n = 1038, 9.4%) and cancer (n = 1013, 9.1%). Corticosteroid use ranged from 11.5% to 64.9% in the dementia and pulmonary subgroups, respectively. Piperacillin-tazobactam use ranged from 9.1% to 28.0% in the pulmonary and cancer subgroups, respectively. The use of thoracic computed tomography ranged from 5.7% to 36.3% in the dementia and cancer subgroups, respectively. Adjusting for patient factors, the risk of in-hospital death was greater in the cancer (adjusted odds ratio [OR] 3.12, 95% confidence interval [CI] 2.44-3.99), dementia (adjusted OR 1.57, 95% CI 1.05-2.35), heart failure (adjusted OR 1.66, 95% CI 1.35-2.03) and DM-HF-Pulm subgroups (adjusted OR 1.35, 95% CI 1.12-1.61), and lower in the diabetes subgroup (adjusted OR 0.67, 95% CI 0.50-0.89), compared with the low comorbidity group. INTERPRETATION: Patients admitted to hospital with CAP cluster into clinically recognizable subgroups based on coexisting conditions. Clinical care and outcomes vary among these subgroups with little evidence to guide decision-making, highlighting opportunities for research to personalize care.

Automated identification of unstandardized medication data: a scalable and flexible data standardization pipeline using RxNorm on GEMINI multicenter hospital data.

Objective: Patient data repositories often assemble medication data from multiple sources, necessitating standardization prior to analysis. We implemented and evaluated a medication standardization procedure for use with a wide range of pharmacy data inputs across all drug categories, which supports research queries at multiple levels of granularity. Methods: The GEMINI-RxNorm system automates the use of multiple RxNorm tools in tandem with other datasets to identify drug concepts from pharmacy orders. GEMINI-RxNorm was used to process 2 090 155 pharmacy orders from 245 258 hospitalizations between 2010 and 2017 at 7 hospitals in Ontario, Canada. The GEMINI-RxNorm system matches drug-identifying information from pharmacy data (including free-text fields) to RxNorm concept identifiers. A user interface allows researchers to search for drug terms and returns the relevant original pharmacy data through the matched RxNorm concepts. Users can then manually validate the predicted matches and discard false positives. We designed the system to maximize recall (sensitivity) and enable excellent precision (positive predictive value) with efficient manual validation. We compared the performance of this system to manual coding (by a physician and pharmacist) of 13 medication classes. Results: Manual coding was performed for 1 948 817 pharmacy orders and GEMINI-RxNorm successfully returned 1 941 389 (99.6%) orders. Recall was greater than 0.985 in all 13 drug classes, and the F1-score and precision remained above 0.90 in all drug classes, facilitating efficient manual review to achieve 100% precision. GEMINI-RxNorm saved time substantially compared with manual standardization, reducing the time taken to review a pharmacy order row from an estimated 30 to 5 s and reducing the number of rows needed to be reviewed by up to 99.99%. Discussion and Conclusion: GEMINI-RxNorm presents a novel combination of RxNorm tools and other datasets to enable accurate, efficient, flexible, and scalable standardization of pharmacy data. By facilitating efficient manual validation, the GEMINI-RxNorm system can allow researchers to achieve near-perfect accuracy in medication data standardization.

Physician antimicrobial prescribing and patient outcomes on general medical wards: a multicentre retrospective cohort study.

BACKGROUND: Variability in antimicrobial prescribing may indicate an opportunity for improvement in antimicrobial use. We sought to measure physician-level antimicrobial prescribing in adult general medical wards, assess the contribution of patient-level factors to antimicrobial prescribing and evaluate the association between antimicrobial prescribing and clinical outcomes. METHODS: Using the General Medicine Inpatient Initiative (GEMINI) database, we conducted a retrospective cohort study of physician-level volume and spectrum of antimicrobial prescribing in adult general medical wards in 4 academic teaching hospitals in Toronto, Ontario, between April 2010 and December 2019. We stratified physicians into quartiles by hospital site based on volume of antimicrobial prescribing (days of therapy per 100 patient-days and antimicrobial-free days) and antibacterial spectrum (modified spectrum score). The modified spectrum score assigns a value to each antibacterial agent based on the breadth of coverage. We assessed patient-level differences among physician quartiles using age, sex, Laboratory-based Acute Physiology Score, discharge diagnosis and Charlson Comorbidity Index. We evaluated the association of clinical outcomes (in-hospital 30-day mortality, length of stay, intensive care unit [ICU] transfer and hospital readmission) with antimicrobial volume and spectrum using multilevel modelling. RESULTS: The cohort consisted of 124 physicians responsible for 124 158 hospital admissions. The median physician-level volume of antimicrobial prescribing was 56.1 (interquartile range 51.7-67.5) days of therapy per 100 patient-days. We did not find any differences in baseline patient characteristics by physician prescribing quartile. The difference in mean prescribing between quartile 4 and quartile 1 was 15.8 days of therapy per 100 patient-days (95% confidence interval [CI] 9.6-22.0), representing 30% higher antimicrobial prescribing in the fourth quartile than the first quartile. Patient in-hospital deaths, length of stay, ICU transfer and hospital readmission did not differ by physician quartile. In-hospital mortality was higher among patients cared for by prescribers with higher modified spectrum scores (odds ratio 1.13, 95% CI 1.04-1.24). INTERPRETATION: We found that physician-level variability in antimicrobial prescribing was not associated with differences in patient characteristics or outcomes in academic general medicine wards. These findings provide support for considering the lowest quartile of physician antimicrobial prescribing within each hospital as a target for antimicrobial stewardship.

Using the Hospital Frailty Risk Score to assess mortality risk in older medical patients admitted to the intensive care unit.

BACKGROUND: Prognostic information at the time of hospital discharge can help guide goals-of-care discussions for future care. We sought to assess the association between the Hospital Frailty Risk Score (HFRS), which may highlight patients' risk of adverse outcomes at the time of hospital discharge, and in-hospital death among patients admitted to the intensive care unit (ICU) within 12 months of a previous hospital discharge. METHODS: We conducted a multicentre retrospective cohort study that included patients aged 75 years or older admitted at least twice over a 12-month period to the general medicine service at 7 academic centres and large community-based teaching hospitals in Toronto and Mississauga, Ontario, Canada, from Apr. 1, 2010, to Dec. 31, 2019. The HFRS (categorized as low, moderate or high frailty risk) was calculated at the time of discharge from the first hospital admission. Outcomes included ICU admission and death during the second hospital admission. RESULTS: The cohort included 22 178 patients, of whom 1767 (8.0%) were categorized as having high frailty risk, 9464 (42.7%) as having moderate frailty risk, and 10 947 (49.4%) as having low frailty risk. One hundred patients (5.7%) with high frailty risk were admitted to the ICU, compared to 566 (6.0%) of those with moderate risk and 790 (7.2%) of those with low risk. After adjustment for age, sex, hospital, day of admission, time of admission and Laboratory-based Acute Physiology Score, the odds of ICU admission were not significantly different for patients with high (adjusted odds ratio [OR] 0.99, 95% confidence interval [CI] 0.78 to 1.23) or moderate (adjusted OR 0.97, 95% CI 0.86 to 1.09) frailty risk compared to those with low frailty risk. Among patients admitted to the ICU, 75 (75.0%) of those with high frailty risk died, compared to 317 (56.0%) of those with moderate risk and 416 (52.7%) of those with low risk. After multivariable adjustment, the risk of death after ICU admission was higher for patients with high frailty risk than for those with low frailty risk (adjusted OR 2.86, 95% CI 1.77 to 4.77). INTERPRETATION: Among patients readmitted to hospital within 12 months, patients with high frailty risk were similarly likely as those with lower frailty risk to be admitted to the ICU but were more likely to die if admitted to ICU. The HFRS at hospital discharge can inform prognosis, which can help guide discussions for preferences for ICU care during future hospital stays.

Data-driven approach to identifying potential laboratory overuse in general internal medicine (GIM) inpatients.

BACKGROUND: Reducing laboratory test overuse is important for high quality, patient-centred care. Identifying priorities to reduce low value testing remains a challenge. OBJECTIVE: To develop a simple, data-driven approach to identify potential sources of laboratory overuse by combining the total cost, proportion of abnormal results and physician-level variation in use of laboratory tests. DESIGN, SETTING AND PARTICIPANTS: A multicentre, retrospective study at three academic hospitals in Toronto, Canada. All general internal medicine (GIM) hospitalisations between 1 April 2010 and 31 October 2017. RESULTS: There were 106 813 GIM hospitalisations during the study period, with median hospital length-of-stay of 4.6 days (IQR: 2.33-9.19). There were 21 tests which had a cumulative cost >US$15 400 at all three sites. The costliest test was plasma electrolytes (US$4 907 775), the test with the lowest proportion of abnormal results was red cell folate (0.2%) and the test with the greatest physician-level variation in use was antiphospholipid antibodies (coefficient of variation 3.08). The five tests with the highest cumulative rank based on greatest cost, lowest proportion of abnormal results and highest physician-level variation were: (1) lactate, (2) antiphospholipid antibodies, (3) magnesium, (4) troponin and (5) partial thromboplastin time. In addition, this method identified unique tests that may be a potential source of laboratory overuse at each hospital. CONCLUSIONS: A simple multidimensional, data-driven approach combining cost, proportion of abnormal results and physician-level variation can inform interventions to reduce laboratory test overuse. Reducing low value laboratory testing is important to promote high value, patient-centred care.

Temporal Clustering of Critical Illness Events on Medical Wards.

Importance: Recognizing and preventing patient deterioration is important for hospital safety. Objective: To investigate whether critical illness events (in-hospital death or intensive care unit [ICU] transfer) are associated with greater risk of subsequent critical illness events for other patients on the same medical ward. Design, Setting, and Participants: Retrospective cohort study in 5 hospitals in Toronto, Canada, including 118 529 hospitalizations. Patients were admitted to general internal medicine wards between April 1, 2010, and October 31, 2017. Data were analyzed between January 1, 2020, and April 10, 2023. Exposures: Critical illness events (in-hospital death or ICU transfer). Main Outcomes and Measures: The primary outcome was the composite of in-hospital death or ICU transfer. The association between critical illness events on the same ward across 6-hour intervals was studied using discrete-time survival analysis, adjusting for patient and situational factors. The association between critical illness events on different comparable wards in the same hospital was measured as a negative control. Results: The cohort included 118 529 hospitalizations (median age, 72 years [IQR, 56-83 years]; 50.7% male). Death or ICU transfer occurred in 8785 hospitalizations (7.4%). Patients were more likely to experience the primary outcome after exposure to 1 prior event (adjusted odds ratio [AOR], 1.39; 95% CI, 1.30-1.48) and more than 1 prior event (AOR, 1.49; 95% CI, 1.33-1.68) in the prior 6-hour interval compared with no exposure. The exposure was associated with increased odds of subsequent ICU transfer (1 event: AOR, 1.67; 95% CI, 1.54-1.81; >1 event: AOR, 2.05; 95% CI, 1.79-2.36) but not death alone (1 event: AOR, 1.08; 95% CI, 0.97-1.19; >1 event: AOR, 0.88; 95% CI, 0.71-1.09). There was no significant association between critical illness events on different wards within the same hospital. Conclusions and Relevance: Findings of this cohort study suggest that patients are more likely to be transferred to the ICU in the hours after another patient's critical illness event on the same ward. This phenomenon could have several explanations, including increased recognition of critical illness and preemptive ICU transfers, resource diversion to the first event, or fluctuations in ward or ICU capacity. Patient safety may be improved by better understanding the clustering of ICU transfers on medical wards.

An Update to the Kaiser Permanente Inpatient Risk Adjustment Methodology Accurately Predicts In-Hospital Mortality: a Retrospective Cohort Study.

BACKGROUND: Methods to accurately predict the risk of in-hospital mortality are important for applications including quality assessment of healthcare institutions and research. OBJECTIVE: To update and validate the Kaiser Permanente inpatient risk adjustment methodology (KP method) to predict in-hospital mortality, using open-source tools to measure comorbidity and diagnosis groups, and removing troponin which is difficult to standardize across modern clinical assays. DESIGN: Retrospective cohort study using electronic health record data from GEMINI. GEMINI is a research collaborative that collects administrative and clinical data from hospital information systems. PARTICIPANTS: Adult general medicine inpatients at 28 hospitals in Ontario, Canada, between April 2010 and December 2022. MAIN MEASURES: The outcome was in-hospital mortality, modeled by diagnosis group using 56 logistic regressions. We compared models with and without troponin as an input to the laboratory-based acute physiology score. We fit and validated the updated method using internal-external cross-validation at 28 hospitals from April 2015 to December 2022. KEY RESULTS: In 938,103 hospitalizations with 7.2% in-hospital mortality, the updated KP method accurately predicted the risk of mortality. The c-statistic at the median hospital was 0.866 (see Fig. 3) (25th-75th 0.848-0.876, range 0.816-0.927) and calibration was strong for nearly all patients at all hospitals. The 95th percentile absolute difference between predicted and observed probabilities was 0.038 at the median hospital (25th-75th 0.024-0.057, range 0.006-0.118). Model performance was very similar with and without troponin in a subset of 7 hospitals, and performance was similar with and without troponin for patients hospitalized for heart failure and acute myocardial infarction. CONCLUSIONS: An update to the KP method accurately predicted in-hospital mortality for general medicine inpatients in 28 hospitals in Ontario, Canada. This updated method can be implemented in a wider range of settings using common open-source tools.

Comparison of Medical and Mental Health Sequelae Following Hospitalization for COVID-19, Influenza, and Sepsis.

Importance: People who survive hospitalization for COVID-19 are at risk for developing new cardiovascular, neurological, mental health, and inflammatory autoimmune conditions. It is unclear how posthospitalization risks for COVID-19 compare with those for other serious infectious illnesses. Objective: To compare risks of incident cardiovascular, neurological, and mental health conditions and rheumatoid arthritis in 1 year following COVID-19 hospitalization against 3 comparator groups: prepandemic hospitalization for influenza and hospitalization for sepsis before and during the COVID-19 pandemic. Design, Setting, and Participants: This population-based cohort study included all adults hospitalized for COVID-19 between April 1, 2020, and October 31, 2021, historical comparator groups of people hospitalized for influenza or sepsis, and a contemporary comparator group of people hospitalized for sepsis in Ontario, Canada. Exposure: Hospitalization for COVID-19, influenza, or sepsis. Main Outcome and Measures: New occurrence of 13 prespecified conditions, including cardiovascular, neurological, and mental health conditions and rheumatoid arthritis, within 1 year of hospitalization. Results: Of 379 366 included adults (median [IQR] age, 75 [63-85] years; 54% female), there were 26 499 people who survived hospitalization for COVID-19, 299 989 historical controls (17 516 for influenza and 282 473 for sepsis), and 52 878 contemporary controls hospitalized for sepsis. Hospitalization for COVID-19 was associated with an increased 1-year risk of venous thromboembolic disease compared with influenza (adjusted hazard ratio, 1.77; 95% CI, 1.36-2.31) but with no increased risks of developing selected ischemic and nonischemic cerebrovascular and cardiovascular disorders, neurological disorders, rheumatoid arthritis, or mental health conditions compared with influenza or sepsis cohorts. Conclusions and Relevance: In this cohort study, apart from an elevated risk of venous thromboembolism within 1 year, the burden of postacute medical and mental health conditions among those who survived hospitalization for COVID-19 was comparable with other acute infectious illnesses. This suggests that many of the postacute consequences of COVID-19 may be related to the severity of infectious illness necessitating hospitalization rather than being direct consequences of infection with SARS-CoV-2.

Developing and Validating a Prediction Model For Death or Critical Illness in Hospitalized Adults, an Opportunity for Human-Computer Collaboration.

Hospital early warning systems that use machine learning (ML) to predict clinical deterioration are increasingly being used to aid clinical decision-making. However, it is not known how ML predictions complement physician and nurse judgment. Our objective was to train and validate a ML model to predict patient deterioration and compare model predictions with real-world physician and nurse predictions. DESIGN: Retrospective and prospective cohort study. SETTING: Academic tertiary care hospital. PATIENTS: Adult general internal medicine hospitalizations. MEASUREMENTS AND MAIN RESULTS: We developed and validated a neural network model to predict in-hospital death and ICU admission in 23,528 hospitalizations between April 2011 and April 2019. We then compared model predictions with 3,374 prospectively collected predictions from nurses, residents, and attending physicians about their own patients in 960 hospitalizations between April 30, and August 28, 2019. ML model predictions achieved clinician-level accuracy for predicting ICU admission or death (ML median F1 score 0.32 [interquartile range (IQR) 0.30-0.34], AUC 0.77 [IQ 0.76-0.78]; clinicians median F1-score 0.33 [IQR 0.30-0.35], AUC 0.64 [IQR 0.63-0.66]). ML predictions were more accurate than clinicians for ICU admission. Of all ICU admissions and deaths, 36% occurred in hospitalizations where the model and clinicians disagreed. Combining human and model predictions detected 49% of clinical deterioration events, improving sensitivity by 16% compared with clinicians alone and 24% compared with the model alone while maintaining a positive predictive value of 33%, thus keeping false alarms at a clinically acceptable level. CONCLUSIONS: ML models can complement clinician judgment to predict clinical deterioration in hospital. These findings demonstrate important opportunities for human-computer collaboration to improve prognostication and personalized medicine in hospital.